Trends in incidence for the main types of childhood cancer are analysed using data from the British National Registry of Childhood Tumours; published reports from other registries are summarized. For registries included in Cancer Incidence in Five Continents, changes in rates reported in successive volumes are calculated and a number of problems in the interpretation of these results are identified. There is some limited evidence for a small increase in the incidence of ALL and more evidence for an increase in brain tumours, although the latter may be at least partly due to changes in diagnostic practice. There is limited, but good, evidence for an increase in the incidence of neuroblastoma. With two exceptions, there is no good evidence for any large increase in incidence for any form of childhood cancer in Britain or other countries over the past 20-30 years. The first exception is Kaposi's sarcoma in Uganda, the cumulative incidence in childhood increasing from fewer than 40 per million in 1968-1982 to more than 500 per million in 1989-1991. The second exception is thyroid carcinoma in Belarus, where the reported incidence probably represents about a 50-fold increase--to about 80 per million per year. This may, however, be at least partly a consequence of improved ascertainment following a screening programme. The results presented here suggest that no widespread important risk factor has been introduced over the past 30 years--unless some other similar factor has been simultaneously removed. Analyses of trends in mortality for childhood cancers are now mainly useful as a means of showing the impact of improved treatment methods.